The traffic from ChatGPT towards the main online stores has increased by almost nine times in one year: +799%. This is the data that best tells you what's happening at the first step of the purchase.
Be careful how you read it: nine times almost nothing remains little. But the direction is clear, and those who sell have a rare advantage: the ability to prepare before the channel becomes large.

What changes compared to Google
On a search engine, the game is about ranking: ten results, you're fighting for first place. In a conversation with an assistant, there aren't ten positions. There's one answer, with two or three names in it.

This reverses two habits.
- It's not enough to be findable, you have to be summarizable. The assistant reports what it can reliably extract: numbers, materials, measurements, return conditions. The mood sentences don't survive the summary.
- The sources cited matter more than the keywords. If a data sheet is in an unreadable PDF or inside an image, it doesn't exist for that reader.
The three questions your page must answer itself

Someone who asks an assistant for advice doesn't write "wood cutting board." They write entire sentences: «"I'm looking for a wooden cutting board for a 40cm countertop that won't crack in the dishwasher, under 50 euros."».
For your page to appear in that response, it must clearly contain:
- the real measure, in centimeters, written in the text;
- the limit of use, explicitly stated («not dishwasher safe» is helpful information, not a weakness);
- the price or range, at a stable point on the page.
These are the same three things that reduce returns. The work done for AI here is the same work that serves the undecided human customer.
Measure it, instead of believing it
Traffic from chatbots is seen in statistics tools like referral, with recognizable domains (`chatgpt.com`, `perplexity.ai`, and the like). It's worth checking it once a month and noting two numbers: how many visits arrive and how long they stay.
The second number is the interesting one. Anyone who comes to an assistant has already made the comparison: either it's the right page, and then it stays, or they realized in three seconds that the assistant gave the wrong advice, and that's your problem, because it means the page promises something it doesn't deliver.
It's not a channel to conquer, it's an extra reader

The temptation, when faced with a +799%, is to rush to "optimize for AI" by buying something. It's early and there's no point.
Only one thing is needed, and there are no shortcuts: the actual product information must be written, consistent, and in the same language customers ask. Those who already have it can easily access this channel. Those whose product descriptions are made up of adjectives cannot.
The rest, understanding whether it's worth monitoring, and with what content, will be decided in six months, with your numbers in hand and not those of a press release.
If you want to understand where your listing traffic is coming from today, write to me.
Sources
- Food Affairs, AI is making retail sites soar: ChatGPT has seen a traffic boom on major e-commerce sites., March 12, 2026, change in referral traffic to retailers (+799% year-on-year)
- NetRetail 2026: Average number of touchpoints activated before purchase
The data concerns the main digital retailers and measures a percentage change on low starting values: it should be read as a trend, not as absolute volume.